Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in
AI Snake Oil Arvind Narayanan
AI companies currently earn revenue primarily from inference charges, but economic theory and historical analysis suggest this business model faces a commodity trap where undifferentiated products and low switching costs will force prices toward marginal cost. AI labs are already moving up the stack through vertical integration, embedded enterprise solutions, and lock-in strategies borrowed from enterprise software to escape this trap, but this shift raises concerns about competition and customer lock-in. The industry's long-term profitability will likely depend on whether companies can replicate software-like properties or achieve market concentration, rather than relying on infrastructure-layer margins that have historically compressed across railroads, telecom, and other capital-intensive industries.
Why it matters
Critics and boosters are both looking in the wrong place
